Linear signaling in the Toll-Dorsal pathway of Drosophila: activated Pelle kinase specifies all threshold outputs of gene expression while the bHLH protein Twist specifies a subset

Dorsal is distributed in a broad concentration gradient in the precellular Drosophila embryo (reviewed by Belvin and Anderson, 1996; Rusch and Levine, 1996). It establishes dorsoventral patterning by regulating a number of target genes in a concentration-dependent fashion. At least five thresholds of gene activity have been described (Huang et al., 1997). Peak levels of Dorsal are required for the activation of the snail gene in the presumptive mesoderm in ventral regions of the early embryo (Ip et al., 1992a), whereas progressively lower levels of Dorsal activate single-minded (sim), ventral nervous system defective (vnd), rhomboid (rho) and short gastrulation (sog) in the neurogenic ectoderm (Kasai et al., 1998; Ip et al., 1992b; von Ohlen and Doe, 2000; Francois et al., 1994). These differential patterns of gene activity subdivide the embryo into three primary tissues: mesoderm, neurogenic ectoderm and dorsal ectoderm.

The restricted activation of the Toll receptor results from the localized processing of the Spätzle (Spz) ligand in ventral regions of the precellular embryo (Mizuguchi et al., 1998; Sen et al., 1998). Toll signaling is transduced by the Pelle kinase, which causes the direct or indirect phosphorylation and degradation of Cactus, so that Dorsal is released from the cytoplasm and enters nuclei (Grosshans et al., 1994; Galindo et al., 1995; Towb et al., 1998). A constitutively activated form of Pelle was created by fusing the Pelle kinase catalytic domain to the signal peptide, the extracellular domain and the transmembrane peptide of a constitutively activated form of the Torso (Tor) receptor, Tor⁴⁰²¹ (Y327C) (Sprenger and Nüsslein-Volhard, 1992). The resulting ‘Pelle-Tor⁴⁰²¹’ chimera caused a dominant ventralized phenotype when its RNA was injected throughout the embryo in a manner independent of both Toll receptor and Tube adapter protein (Grosshans et al., 1994; Galindo et al., 1995). This is consistent with the idea that activation of Pelle leads to efficient degradation of Cactus, so that Dorsal can enter nuclei in both dorsal and ventral regions of the mutant embryo. However, mutant phenotypes were assessed predominantly on the basis of cuticular defects. Consequently, it is not known whether physiological levels of activated Pelle kinase are sufficient for the activation of type I target genes, such as twist and snail, which depend on peak concentrations of the Dorsal gradient (Huang et al., 1997). It is conceivable that the Toll-Dorsal signaling pathway is branched. For example, activation of the Toll receptor might induce Pelle and an additional kinase, which together activate type I target genes (e.g. Karin, 1999).

twist is one of the first genes activated by the Dorsal gradient (Thisse et al., 1991; Jiang et al., 1991). It encodes a bHLH regulatory protein implicated in mesoderm differentiation in a broad spectrum of animals (Reuter and Leptin, 1994; Baylies and Bate, 1996; Harfe et al., 1998). This conservation of Twist contrasts with the apparently specific use of Dorsal as a dorsoventral determinant in insect embryos (Chen et al., 2000). Thus far, Rel-containing transcription factors have not been implicated in the dorsoventral patterning of vertebrate embryos, even though fly and frog embryos employ many common signaling components such as transforming growth factor β (TGFβ) and Chordin/Sog inhibitors (Ferguson, 1996).

Linearity in the Toll-Dorsal signaling pathway was analyzed by creating ectopic anterior-posterior gradients of Twist and activated Pelle kinases (Pelle-Tor⁴⁰²¹ and Pelle-Tor). Twist- and Pelle-coding sequences were attached to the bicoid (bcd) 3′UTR and expressed in the maternal germline using the hsp83 promoter (Huang et al., 1997). The Pelle-Tor⁴⁰²¹ fusion gene is sufficient to establish sequential patterns of snail, sim, vnd and sog expression across the anteroposterior axis of transgenic embryos. These results suggest that the levels of activated Pelle kinase determine dorsoventral patterning thresholds, and argue against branching in the Toll signaling pathway. The twist transgene was able to activate snail, sim and vnd expression in mutant embryos containing low, uniform levels of the Dorsal protein. However, gene expression supported by Twist in the absence of a Dorsal gradient is erratic, in that expression patterns are out of order or are incorrect. These observations contrast with the recent demonstration that most Bicoid gradient thresholds are generated also by Hunchback, an immediate target of the Bicoid activator. As Twist largely fails to compensate for the loss of the Dorsal gradient, we conclude that Dorsal and Twist function in a highly interdependent manner to specify the mesoderm and ventral regions of the neurogenic ectoderm.

The construction of Pelle-Tor⁴⁰²¹ and Pelle-Tor fusion proteins in pSELECT vector (Promega) have been described previously (Galindo et al., 1995). The chimeric proteins contain the extracellular and transmembrane domains of Tor and Tor⁴⁰²¹ residues 1-455 fused to residues 163-501 of Pelle. To generate anteroposterior Pelle-Tor and Pelle-Tor⁴⁰²¹ gradients, we placed these fusions under the control of the hsp83 promoter and regulation by the bcd 3′UTR as follows. Blunted-NcoI/SacI fragments containing the Pelle-Tor- and Pelle-Tor⁴⁰²¹-coding sequences were cloned into the blunted-HindIII/SacI digested hsp83/pUC19 (Huang et al., 1997). A three-way ligation of a KpnI/blunted-SacI digested fragment containing hsp83-Pelle-Tor or hsp83-Pelle-Tor⁴⁰²¹, blunted-SacII/XbaI fragment containing the bcd 3′UTR, and KpnI/XbaI-digested Casper AUG β-gal P-element transformation vector (Huang et al., 1997) generated the Pelle-Tor⁴⁰²¹ and Pelle-Tor P-element transformation vectors used in this study. A blunted HindIII/HincII fragment containing the entire twist open reading frame (ORF) and a 5′ β-globin leader was isolated from the Twist/pNB40 plasmid (kindly provided by Maria Leptin) and cloned into the EcoRV site of pBSK+-KpnII, creating Twist-KpnII/SK+. pBSK+-KpnII is pBSK+ (Stratagene), which was modified to contain an additional KpnI site in place of SacI. A KpnI fragment containing the twist ORF from twist-KpnII/SK+ was inserted into the KpnI site of the pCASPER-hsp83-KpnI-bcd3′UTR vector (a modified form of pCAS-AUGβgal obtained from Hailan Zhang). twist was thereby placed under the control of the hsp83 maternal promoter and bcd3′UTR localization sequence to generate P-element transformation vector P(Twist-bcd). The construction of P(bcdToll^10bbcd) and P(hspToll^10bbcd) P-element transformation vectors have been previously described (Huang et al., 1997) which places the Toll^10b sequence under control of the bcd 3′UTR localization sequence and the bcd and hsp83 promoters, respectively. The bcd promoter mediates maternal expression at levels lower than the hsp83 maternal promoter. P-element mediated transformation and in situ hybridization using digoxigenin-labeled RNA probes were performed as described previously (Jiang et al., 1991).

All mutant fly stocks were obtained from the Bloomington Stock Center (Indiana) and wild-type embryos correspond to the yw fly stock, unless otherwise noted. The ectopic Toll^10b, Pelle-Tor⁴⁰²¹, Pelle-Tor and Twist-bcd anteroposterior Dorsal gradients were examined in the absence of the endogenous dorsoventral Dorsal gradient by introducing the Toll^10b, Pelle-Tor⁴⁰²¹, Pelle-Tor and Twist-bcd transgenes into embryos homozygous for a null mutation in gastrulation defective (gd7) (Konrad et al., 1988). Male flies from several independent lines carrying these P-element transposons were individually crossed to gd7/FM3 females. gd⁷;(P-element)/+ males were crossed to gd⁷/FM3 females and embryos from gd⁷/gd⁷; (P-element)/+ females were collected for analysis by in situ hybridization.

The ectopic Toll^10b anteroposterior Dorsal gradient was examined in the absence of Twist by crossing the P(hspToll^10bbcd) transgene into a twist-/twist- mutant background. Sp/CyO; P(hspToll^10bbcd)/+ males were crossed to cn twi¹ bw/CyO females obtained from Bloomington stock center. Virgin females of the genotype cn twi¹ bw/CyO; P(hspToll^10bbcd) were mated to cn twi¹ bw/CyO males. twist-/twist- mutant embryos represented one fourth of the embryos examined. Similar crosses were done to introduce the P(Twist-bcd) transgene into a snail mutant background (yw; snail^IIG05/CyO ftz-lacZ; kindly provided by Tony Ip).

The ectopic Twist anteroposterior gradient was examined in a background of low uniform Dorsal levels by misexpressing the twist-bcd transgene in embryos homozygous for a partially activating mutation of Toll (Schneider et al., 1991). Male flies from several independent lines in which the P(Twist-bcd) insertion was mapped to the second chromosome and balanced by CyO were individually crossed to Sp/CyO;Toll^rm9/TM3 females, which were isolated by crossing Toll^rm9/TM3 males to double balancer Sp/CyO; PrDr/TM3 females. twist-bcd/CyO; Toll^rm9/TM3 males were crossed to Toll^rm10/TM3 females, and embryos from twist-bcd/+; Toll^rm9/Toll^rm10 females were collected from at least three individual lines for analysis by in situ hybridization.

Flies were allowed to deposit eggs on fresh apple juice agar plates with yeast paste for 2 hours, and then removed. Embryos were collected after aging an additional 2 hours at room temperature and homogenized in RIPA Buffer [50 mM TrisHCl (pH 8.0), 150 mM NaCl, 1% NP-40, 0.5% deoxycholate, 0.1% SDS and protease inhibitors]. Extracts were centrifuged at 16,000 g for 15 minutes. The clear protein supernatant was carefully separated from floating lipid and precipitate, and quantified by Bradford assay. Equivalent amounts of embryo extracts were subjected to SDS-PAGE and western blotting using an affinity-purified, rabbit polyclonal antibody to Pelle protein at a 1:2000 dilution (kindly provided by James Manley) (Shen and Manley, 1998).

The snail, sim, vnd and sog expression patterns represent four different Toll-Dorsal signaling thresholds. snail is activated only by peak levels of the Dorsal gradient (Ip et al., 1992a), sim and vnd are activated by intermediate levels (Kasai et al., 1998; von Ohlen and Doe, 2000), and sog is activated by the lowest levels of the gradient (Francois et al., 1994). These expression patterns were visualized in mutant and transgenic embryos via in situ hybridization using digoxigenin-labeled antisense RNA probes (Huang et al., 1997).

Dorsal target genes are essentially silent in mutant embryos that lack an endogenous dorsoventral Dorsal nuclear gradient (Fig. 1A-C). Mutant embryos were collected from females that are homozygous for a null mutation in the gastrulation defective (gd) gene, which blocks the processing of the Spätzle ligand and the activation of the Toll receptor. These mutants permit the analysis of ectopic, anteroposterior Dorsal and Twist gradients in ‘apolar’ embryos that lack dorsoventral polarity. snail (Fig. 1D), vnd, (Fig. 1E) and sog (Fig. 1F) are sequentially expressed along the anteroposterior axis of mutant embryos that contain a constitutively activated form of the Toll receptor (Toll^10b) misexpressed at the anterior pole using the bicoid (bcd) promoter and 3′ UTR (Fig. 1). These expression patterns depend on an ectopic anteroposterior Dorsal nuclear gradient (Huang et al., 1997). The repression of the vnd and sog patterns at the anterior pole is probably mediated by Snail, which normally excludes expression of these genes in the ventral mesoderm of wild-type embryos (Mellerick and Nirenberg, 1995; Rusch and Levine, 1996).

The activated Pelle-Tor⁴⁰²¹ kinase also directs sequential anteroposterior patterns of snail, vnd, and sog expression in gd^–/gd^– mutant embryos (Fig. 2C-E). As in the case of Toll^10b, the activated Pelle kinase was misexpressed at the pole using the bcd 3′ UTR. The snail, vnd and sog expression patterns are similar to those obtained with the Toll^10b transgene (see Fig. 1D-F). The vnd and sog expression patterns are probably repressed at the anterior pole by Snail (Fig. 2D,E). These results suggest that the levels of Pelle kinase activity are sufficient to determine different Dorsal transcription thresholds.

Similar experiments were carried out with a Pelle-Tor fusion gene that contains the Tor signal peptide, extracellular domain and transmembrane peptide, but lacks the amino acid substitution (Y327C) in the Tor⁴⁰²¹ protein that induces dimerization (Fig. 2F-H). The Pelle-Tor fusion protein fails to induce snail expression (Fig. 2F), but succeeds in activating vnd (Fig. 2G) and sog (Fig. 2H). The activities of the Pelle-Tor⁴⁰²¹ and Pelle-Tor proteins are summarized in Fig. 2B.

Western assays were carried out to determine whether the different activities of Pelle-Tor⁴⁰²¹ and Pelle-Tor might result from the differential stability of the two fusion proteins (Fig. 2A). The normal Pelle kinase has a molecular weight of ∼75 kDa (lane 1, Fig. 2A), whereas the Pelle-Tor⁴⁰²¹ and Pelle-Tor fusion proteins are considerably larger, ∼140 kDa (lanes 2 and 3, respectively). The Pelle-Tor fusion protein (lane 3) is expressed at somewhat higher levels than the Pelle-Tor⁴⁰²¹ protein, but nonetheless fails to activate snail expression. One interpretation of these results is that recruitment to the plasma membrane is not sufficient for the induction of peak Pelle kinase activity. Rather, full induction might require both recruitment and protein dimerization, which is achieved with the PelleTor⁴⁰²¹ fusion protein, but not with Pelle-Tor (Grosshans et al., 1994; Galindo et al., 1995). The PelleTor⁴⁰²¹ protein is constitutively activated in the absence of the Tor ligand, whereas full activation of Pelle-Tor might require ligand (see Discussion).

There are several similarities between Dorsal and the major determinant of anteroposterior patterning, Bicoid, even though the two morphogens are unrelated. Both gradients activate regulatory genes that are essential for patterning the early embryo. Bicoid activates the zinc finger gene Hunchback, while Dorsal activates the bHLH gene Twist. It has recently been shown that the loss of the normal Bicoid gradient can be largely compensated by an anteroposterior Hunchback gradient (Wimmer et al., 2000). Hunchback restores posterior head segments and the thoracic segments lost in bicoid^–/bicoid^– mutant embryos. We have investigated the possibility that Twist plays a similar role in dorsoventral patterning.

Dorsoventral patterning genes exhibit abnormal patterns of expression in twist^–/twist^– mutant embryos (Fig. 3). For example, snail expression is reduced, and the residual snail pattern exhibits periodicity along the anteroposterior axis (Fig. 3C,D; compare with 3A,B). The activities of the constitutively activated Toll^10b receptor are significantly impaired in twist^–/twist^– mutant embryos (Fig. 3E,F). This experiment involved the use of an hsp83-Toll^10b-3′ bcd UTR transgene, which induces a broad anteroposterior Dorsal nuclear gradient (Huang et al., 1997). This transgene is similar to the one used in Fig. 1D-F, except that Toll^10b was expressed using the stronger hsp83 promoter rather than the bicoid promoter. In an otherwise normal genetic background, the Toll^10b transgene leads to the ectopic activation of snail expression in the presumptive head and anterior thorax (Fig. 3E). However, in twist^–/twist^– embryos there is reduced expression of both the endogenous snail expression pattern, and the ectopic pattern in anterior regions (Fig. 3F). Thus, even abnormally high levels of Toll signaling cannot compensate for the loss of Twist. The gap in snail expression seen at the boundary between the ectopic and endogenous patterns (arrow, Fig. 3E) may be due to an unknown repressor that is regulated by low levels of the Dorsal gradient. This gap is more pronounced in twist^–/twist^– mutants (arrow, Fig. 3F), suggesting that Twist is not required for the activation of the putative snail repressor.

The preceding results suggest that Twist is necessary for specifying Dorsal gradient thresholds. To determine whether Twist might be sufficient, a twist-bcd transgene was introduced into a number of genetic backgrounds. In wild-type embryos, the twist-bcd transgene induces ectopic activation of the sim gene in anterior regions (Fig. 4D; compare with 4A). This activation is more easily visualized in snail^–/snail^– mutant embryos because Snail represses sim expression in the ventral mesoderm (Kasai et al., 1992). There is a delay in the onset of sim expression in snail^–/snail^– embryos (Fig. 4B,C), and staining is sporadic in ventral regions where the gene is normally repressed (Fig. 4C; compare with 4A). However, the twist-bcd transgene, when introduced into snail^–/snail^– mutant embryos, induces strong expression of sim in anterior regions (Fig. 4E,F), although staining is largely restricted to ventral regions where there are high levels of the Dorsal gradient (Fig. 4F).

In addition, females containing the twist-bcd transgene exhibit a daughterless phenotype as only male progeny are observed. We examined Sex-lethal expression in embryos obtained from these females. Sex-lethal is required for sex determination in females, and this gene is expressed in female but not male embryos (Fig. 4H,G, respectively) (Bopp et al., 1991). Sex-lethal expression is repressed in the anterior of embryos obtained from females containing the twist-bcd transgene (Fig. 4I). This result suggests that Twist may form heterodimeric complexes with Daughterless/Scute, two bHLH proteins required for Sex-lethal expression, thereby titrating the levels of these proteins and making them unable to activate Sex-lethal. The ability of Twist to form homodimeric as well as heterodimeric complexes with other bHLH proteins present in the early embryo may regulate its own dorsoventral patterning activities as well (see Discussion).

In order to determine whether Twist patterning activities require a Dorsal gradient, the twist-bcd transgene was introduced into mutant embryos derived from Toll^rm9/Toll^rm10 transheterozygous females. This mutant Toll receptor is defective and partially active in a Spätzle-independent fashion (Schneider et al., 1991). Consequently, mutant embryos contain uniform, low levels of the Dorsal protein in both dorsal and ventral nuclei. The twist-bcd transgene causes a substantial reorganization in the patterning of mutant embryos (Fig. 5).

The low, uniform levels of Dorsal present in Toll^rm9/Toll^rm10 mutant embryos are insufficient to activate snail in precellular or newly cellularized embryos (Fig. 5A). The twist-bcd transgene activates snail throughout anterior regions of precellular embryos (Fig. 5B); however, this staining pattern is refined into a broad anterior stripe after cellularization (Fig. 5C). This ectopic snail expression pattern exhibits a sharp posterior border, suggesting that the Twist gradient is sufficient for the on/off regulation of snail in the absence of a Dorsal gradient (see Discussion).

sog is normally activated throughout the neurogenic ectoderm by the lowest levels of the Dorsal gradient (Francois et al., 1994; Markstein et al., 2002). The low levels of Dorsal present in Toll^rm9/Toll^rm10 mutant embryos are sufficient to activate sog everywhere except the extreme termini (Fig. 5D). The twist-bcd transgene leads to the loss of sog expression in anterior regions (Fig. 5E), probably because of repression by Snail. As mentioned earlier, Snail also appears to repress vnd and sog expression in anterior regions of transgenic embryos that contain the Toll^10b or Pelle-Tor⁴⁰²¹ transgenes (see Figs 1 and 2).

The low levels of Dorsal present in Toll^rm9/Toll^rm10 mutant embryos are insufficient to activate sim, although there is occasional staining in the posterior pole (data not shown). The twist-bcd transgene leads to the efficient activation of sim in anterior regions (Fig. 5F). Staining appears to be restricted to those regions where snail expression is lost (Fig. 5C). These results suggest that a Twist gradient is sufficient to generate multiple dorsoventral patterning thresholds (sim and snail) in the presence of low, uniform levels of Dorsal.

The twist-bcd transgene was introduced into mutant embryos that completely lack Dorsal (Fig. 6). Without the transgene these mutants do not express twist, snail, sim, vnd or sog (Fig. 6A; data not shown). Introduction of the twist-bcd transgene causes intense expression of twist in the anterior 40% of the embryo (Fig. 6B; compare with 6A). This broad Twist gradient fails to activate snail (data not shown), but succeeds in inducing weak expression of sim (Fig. 6C) and somewhat stronger staining of vnd (Fig. 6D) at the anterior pole. The activation of vnd in mutant embryos is comparable with the expression seen in wild-type and Toll^rm9/Toll^rm10 embryos (Fig. 6E; data not shown). However, in both wild-type and mutant embryos the vnd pattern is transient, and lost after the completion of cellularization (Fig. 6F; data not shown). These results indicate that Twist can activate dorsoventral patterning genes in the absence of Dorsal.

The demonstration that the Pelle-Tor⁴⁰²¹ fusion protein can generate most or all dorsoventral patterning thresholds suggests that the levels of Pelle kinase activity determine Spätzle-Toll signaling thresholds. There is no need to invoke branching in the Toll pathway upstream of Pelle, although it is possible that branching exists downstream of Pelle. For example, there is emerging evidence that IkB kinases might trigger the degradation of Cactus in the immune response (Rutshmann et al., 2000; Silverman et al., 2000; Lu et al., 2001). It is currently unclear whether these kinases are also required for dorsoventral patterning and, if so, whether they are induced by Pelle or separately activated in response to Toll signaling.

Multiple patterning thresholds are also specified by different levels of kinase activity in the Tor, epidermal growth factor receptor (EGFR) and Sevenless (Sev) signaling pathways (Greenwood and Struhl, 1997; Halfar et al., 2001). In the embryo, low levels of Ras1 are sufficient to activate the Tor target gene tailless, whereas higher levels are required for the induction of hückebein (Greenwood and Struhl, 1997). Similarly, in the eye, different levels of EGFR and Sev signaling lead to different levels of MAPK activity: low levels permit the differentiation of R8 photoreceptor cells, whereas higher levels (or more persistent expression) of MAPK activity leads to the specification of an R1-R7 fate (Halfar et al., 2001). In the present study it has been possible to link the levels of an activated cytoplasmic kinase (Pelle) with the expression of well-characterized target genes that are regulated by different concentrations of the Dorsal gradient. It is conceivable that the MAP and Pelle kinases exist in multiple states, which help establish different thresholds of gene activity. For example, even high concentrations of membrane-localized Pelle-Tor fusion protein fail to activate snail. While the Tor⁴⁰²¹ receptor domain of Pelle-Tor⁴⁰²¹ induces ligand-independent dimerization, the wild-type Tor receptor domain contained within the Pelle-Tor fusion might mediate only low levels of dimerization because of competition with the endogenous Tor receptor for binding to the Trunk ligand (Casali and Casanova, 2001). Full activation of the Pelle kinase might depend on recruitment to the plasma membrane and dimerization (Grosshans et al., 1994; Galindo et al., 1995), while recruitment to the membrane alone may produce only partial activation of kinase activity. The induction of full Pelle kinase activity, and ultimately the activation of snail expression, might require trans-phosphorylation induced by dimerization of the Toll receptor (Shen and Manley, 1998). In the context of the normal Toll-Dorsal signaling pathway, the transition between a partially activated Pelle kinase and a fully activated kinase might help generate a sharp on/off pattern of snail expression. Perhaps the Pelle kinase is converted into the fully activated form (required for snail expression) when Toll signaling exceeds a certain minimal threshold.

An anteroposterior Twist gradient generates at least two thresholds of gene activity in mutant embryos that contain decreased levels of Dorsal (summarized in Fig. 7B). High levels of Twist activate sim at the anterior pole, whereas lower levels are sufficient to induce the expression of snail in more posterior regions of embryos containing low, uniform levels of the Dorsal protein (Fig. 5C,F). These results demonstrate that twist gene activity is not dedicated to mesoderm formation. Instead, Twist supports expression of two regulatory genes, sim and vnd, which pattern ventral regions of the neurogenic ectoderm (Crews et al., 1988; McDonald et al., 1998). The twist-bcd transgene was shown to induce weak expression of both genes even in mutant embryos that completely lack Dorsal (Fig. 6C,D).

The Twist gradient exhibits some unexpected activities in Toll^rm9/Toll^rm10 mutant embryos. In particular, both high and intermediate levels of the gradient initially activate snail in a broad domain of precellular embryos (Fig. 5B). However, during cellularization snail is repressed at the anterior pole where there are high levels of Twist (Fig. 5C). Thus, it would appear that high levels of Dorsal + high levels of Twist activate snail expression (e.g. ventral region of wild-type embryos), while low levels of Dorsal + high levels of Twist repress expression (e.g. anterior region of Toll^rm9/Toll^rm10 embryos containing the twist-bcd transgene). The ratio of Dorsal to Twist might keep Twist patterning activity under control, such that high levels of Twist specify mesodermal targets (i.e. snail), lower levels specify neurogenic targets (i.e. sim and vnd) and the expression of genes such as Sxl remain unaffected.

Alternatively, Twist may differentially interact with a number of bHLH proteins that are present in the early embryo (Moore et al., 2000) to affect its patterning activity. For example, Twist is thought to form a heteromeric activation complex with other bHLH proteins including the ubiquitous, maternal bHLH protein Daughterless (Da) (Gonzalez-Crespo and Levine, 1993; Castanon et al., 2001). The loss of Sex-lethal expression at the anterior pole of twist-bcd embryos (Fig. 4I) may result from a failure of Twist-Daughterless heterodimers to activate this gene. It has been demonstrated that Twist-Daughterless heterodimers possess a different patterning activity from Twist-Twist homodimers (Castonon et al., 2001). It is possible that Twist-Daughterless heterodimers formed in twist-bcd embryos actively repress Sex-lethal expression. Alternatively, Twist might titrate Daughterless levels by forming a sterile heterodimeric complex that is not able to activate Sex-lethal. However, Sex-lethal is normally expressed in ventral regions of wild-type embryos that contain both Twist and the ubiquitous Daughterless, therefore regulation of bHLH patterning activities must be more complex. In relation to dorsovental patterning, Twist-Twist complexes might be favored in anterior regions of embryos that express the twist-bcd transgene, while Twist-bHLH complexes are formed away from the anterior pole where expression of the transgene is decreased. These complexes might fail to activate snail, or even actively repress transcription as it has been demonstrated that several bHLH proteins can function as repressors of transcription (Brentrup et al., 2000). Regardless of mechanism, the Twist gradient inverts the order of the sequentially expressed snail and sim genes, when compared with the patterns obtained with the normal Dorsal (and Twist) gradient. In the presence of low, uniform levels of Dorsal, high concentrations of Twist specify mesectoderm (sim expression), while lower levels specify mesoderm (snail expression). These observations raise the possibility of evolutionary plasticity in the use of Twist in tissue specification.

snail is activated by Dorsal and Twist in cellularizing embryos (Ip et al., 1992a). The sharp lateral limits of the snail expression pattern establish the boundary between the presumptive mesoderm and neurogenic ectoderm (Kosman et al., 1991; Leptin, 1991). It has been suggested that the crude Dorsal gradient triggers a somewhat steeper Twist gradient, and the two activators function synergistically within the snail 5′ cis-regulatory DNA to establish the sharp, on/off expression pattern (Ip et al., 1992a). Dorsal-Twist transcription synergy may provide a means for ‘multiplying’ the Dorsal and Twist gradients to produce the sharp snail pattern (see Summary, Fig. 7A). This model suggests that both proteins must be present in a gradient to generate the sharp snail border. However, while both Dorsal and Twist are required for the activation of snail, we have shown that a Twist gradient is sufficient to generate a reasonably sharp pattern of snail expression in embryos containing low, uniform levels of Dorsal. We propose that cooperative binding of Twist might act as a switch to regulate snail expression when the snail 5′ cis-regulatory region is rendered responsive by the Dorsal activator (whether present at uniform levels or in a gradient). Therefore, the ratio of Dorsal to Twist may be important to produce the sharp lateral limits of snail expression.

This study raises some questions about the role of operator binding affinities in the specification of different transcription thresholds. The Dorsal binding sites present in the snail 5′ regulatory region bind with lower affinity than the sites present in the rho lateral stripe enhancer (NEE) (Ip et al., 1992a; Ip et al., 1992b). The analysis of a number of synthetic enhancers prompted the proposal that the activation of Dorsal target genes in the ventral mesoderm versus lateral neurogenic ectoderm depends on the affinity of Dorsal operator sites (Jiang and Levine, 1993; Huang et al., 1997). However, the demonstration that the twist-bcd transgene can activate snail expression in Toll^rm9/Toll^rm10 embryos suggests that occupancy of the distal Dorsal-binding sites may not be crucial for determining whether the gene is on or off. It is conceivable that Dorsal occupies one or more sites in mutant embryos, but is unable to trigger expression in the absence of Twist. In general, ‘promoter context’ (combinations of regulatory factors) might be more critical for defining Dorsal transcription thresholds than the affinities of Dorsal operator sites.

The relationship between Dorsal and Twist appears distinct from the interplay between Bicoid and Hunchback (Wimmer et al., 2000). It has been suggested that the Bicoid gradient is a relatively recent evolutionary innovation for patterning the anterior-posterior axis of long germband insects (Dearden and Akam, 1999). By contrast, Hunchback is ancient and is used in the patterning of short germband insects such as grasshoppers. Most of the patterning activity controlled by the Bicoid gradient appears to be mediated by Hunchback, which is a direct target of the Bicoid activator. Only the patterning of the anteriormost head structures requires Bicoid and cannot be compensated by high levels of Hunchback (Wimmer et al., 2000). Thus, the patterning activity of the Bicoid gradient can be explained by the regulation of several target genes, which is consistent with its recent evolution. By contrast, either Dorsal or Twist protein alone produces only a small subset of the five or six dorsoventral patterning thresholds generated by the concerted action of both proteins. We conclude that Dorsal and Twist work in a highly interdependent and synergistic fashion to regulate a large number of target genes involved in patterning the dorsoventral axis.

We thank Tony Ip for providing snail mutant flies, Maria Leptin for sending a full-length twist cDNA plasmid, James Manley for providing the anti-Pelle antibody, and Steve Wasserman for sending Pelle-Tor and Pelle-Tor⁴⁰²¹ plasmids. This work was funded by a grant from the NIH (GM46638) A. S. is supported by a postdoctoral fellowship from the NIH (GM20352).